As concerns about the spread of political misinformation have mounted, scholars have found that fact-checks can reduce the extent to which people believe misinformation. Whether this finding extends to social media is unclear. Social media is a high-choice environment in which the cognitive effort required to separate truth from fiction, individuals’ penchant for select exposure, and motivated reasoning all may render fact-checks ineffective. Furthermore, large social media companies have largely refrained from permitting external researchers to administer experiments on their platforms. To investigate whether fact-checking can rebut misinformation on social media, we administer two experiments on large, nationally representative samples using a novel platform designed to mimic Facebook’s news feed. We observe corrections having large effects on factual beliefs (.62 on a five-point scale, p<.001𝑝<.001). While misinformation degrades accuracy, our results offer strong evidence that fact-checks increase accuracy, even when tested on realistic simulations of social media platforms.